The Kamchatka result covers one segment on one margin. The published account names the section that ruptured and the locked patch the model had already drawn there [3][8]. It does not report how many patches the algorithm flagged elsewhere along that subduction zone, or how many flagged patches have so far stayed quiet [19]. Without that count, what the test demonstrates is that sparse ground-motion data can resolve an asperity boundary at all, which is the part Periollat says convinced him the approach has potential [9], rather than how often the method points at ground that then does nothing.
That distinction matters because the underlying physics is the model's premise, not its finding. Locked patches store elastic strain and release it when friction gives way; that is the assumption the algorithm encodes when it reads GPS measurements of slow surface movement to find portions of a fault that are stuck [7]. Rupture landing where strain had accumulated is what the theory says should happen. The open question was always the measurement chain, and Kamchatka is one good answer about the measurement chain.
The same region supplies the sharper caution. Kamchatka produced giant earthquakes in 1952 and 2025, seventy-three years apart [10][18], and the recent one generated a much smaller tsunami, which the researchers read as the shallowest part of the fault having slipped less than it did in 1952 [10]. The same location produced two different consequences. The method forecasts where a fault is locked, not what a rupture there will do, and the team is explicit that it predicts neither tsunami size nor timing [11]. For a coastal planner, the wave height still has to come from somewhere else.
That gap points to a geometry problem. The shallow reach of a subduction fault is both the part that drives tsunami and the part that sits offshore, where land GPS stations do not reach and measurements are far scarcer [14]. Researchers in Japan have started measuring slow seafloor deformation with acoustic instruments over multi-year spans, similar work has been proposed for Chile and the Pacific Northwest, and a recently launched satellite could eventually fill in regions with no GPS coverage [15]. Parts of the Pacific remain thin enough on data that tsunami hazard is hard to assess there at all [16]. The method's usefulness at any given margin is going to be set by that instrumentation, not by the algorithm.
The team is now applying it to Japan, Mexico, New Zealand and Cascadia, each with its own complication, including faults that release energy gradually rather than in a single break [12], and is testing whether the same logic separates the creeping and locked stretches of the Hayward Fault in the Bay Area [13]. Funning's own framing is the one to keep: better location forecasting should complement public preparedness, not stand in for it, and where he lives in Southern California he calls a large earthquake a matter of when rather than if [17].
